🔷 Axon

The ML/AI-first systems programming language

Compile-time tensor shapes · Ownership-based safety · Native GPU execution

---

What is Axon?

Axon is a compiled, statically-typed language purpose-built for machine learning and AI. It combines Rust's memory safety with PyTorch's ergonomics and adds compile-time tensor shape checking, so dimension mismatches are caught before your code ever runs.

use std.nn.{Linear, Module};
use std.optim.Adam;
use std.loss.cross_entropy;

model Classifier {
    fc1: Linear<784, 256>,
    fc2: Linear<256, 10>,
}

extend Module for Classifier {
    fn forward(&self, x: Tensor<Float32, [?, 784]>): Tensor<Float32, [?, 10]> {
        val h = relu(self.fc1.forward(x));
        self.fc2.forward(h)
    }
}

fn main() {
    var net = Classifier { fc1: Linear.new(), fc2: Linear.new() };
    var opt = Adam.new(net.parameters(), lr: 0.001);

    for epoch in 0..10 {
        for (images, labels) in &train_loader {
            val loss = cross_entropy(net.forward(images), labels);
            loss.backward();
            opt.step();
            opt.zero_grad();
        }
    }
}

---

Feature Highlights

Feature	Description
🔷 First-class tensors	Tensor<Float32, [?, 784]> — shapes verified at compile time
🔒 Memory safety	Ownership + borrowing — no GC, no data races
🧠 Built-in autograd	Reverse-mode automatic differentiation
🚀 Native GPU	@gpu functions compiled via MLIR to CUDA/ROCm/Vulkan
📐 Shape checker	Matmul, broadcast, reshape errors caught before runtime
🔮 Type inference	Hindley-Milner — write less, know more
📦 Package manager	axonc pkg — create, build, test, publish
🛠️ Rich tooling	LSP, formatter, linter, REPL, doc generator, VS Code extension
🏗️ ML framework	Neural network layers, optimizers, loss functions, data loading

---

Quick Start

Install

# From source
git clone https://github.com/axon-lang/axon.git
cd axon
cargo build --release

Hello, World!

fn main() {
    println("Hello, Axon!");
}

axonc build hello.axon -o hello
./hello

Hello, Tensor!

fn main() {
    val A: Tensor<Float32, [2, 3]> = randn([2, 3]);
    val B: Tensor<Float32, [3, 4]> = randn([3, 4]);
    val C = A @ B;   // Tensor<Float32, [2, 4]> — shape checked!
    println("{}", C);
}

---

Project Structure

axon/
├── src/                    # Compiler source (Rust)
│   ├── lexer.rs            # Lexer
│   ├── parser.rs           # Parser
│   ├── typeck.rs           # Type checker (HM inference)
│   ├── shapes.rs           # Shape checker
│   ├── borrow.rs           # Borrow checker
│   ├── mir.rs              # Mid-level IR
│   ├── codegen/            # LLVM + MLIR backends
│   ├── stdlib/             # Standard library (200+ functions)
│   ├── lsp/                # Language server
│   ├── pkg/                # Package manager
│   ├── fmt.rs              # Formatter
│   ├── lint.rs             # Linter
│   ├── repl.rs             # REPL
│   └── doc.rs              # Doc generator
├── stdlib/                 # Axon source stubs (.axon)
├── tests/                  # 1,034 tests
├── editors/vscode/         # VS Code extension
├── benches/                # Benchmarks
├── fuzz/                   # Fuzz testing
└── docs/                   # Documentation
    ├── guide/              # Language guides
    ├── tutorial/           # Step-by-step tutorials
    ├── reference/          # CLI and error reference
    ├── migration/          # Python/PyTorch migration guides
    └── internals/          # Compiler internals

---

Building from Source

Prerequisites

• Rust 1.75+ (stable)
• Git

Build

git clone https://github.com/axon-lang/axon.git
cd axon
cargo build --release

The compiler binary is at target/release/axonc.

Verify

./target/release/axonc --version
# axonc 0.1.0

---

Running Tests

# Full test suite (1,034 tests)
cargo test

# Specific phases
cargo test --lib lexer           # Lexer tests
cargo test --lib parser          # Parser tests
cargo test --test type_tests     # Type checker tests
cargo test --test codegen_tests  # Code generation tests
cargo test --test stdlib_tests   # Standard library tests
cargo test --test ai_framework_tests  # ML framework tests
cargo test --test tooling_tests  # Tooling tests

---

Documentation

Read the full documentation online

Document	Description
Getting Started	Installation and first project
Language Tour	Quick syntax overview
Tensor Guide	Tensor types, shapes, and operations
GPU Programming	Native GPU support
Ownership & Borrowing	Memory safety model
Error Handling	Option, Result, and ?
Modules & Packages	Module system and Axon.toml
CLI Reference	Complete axonc command reference
Compiler Errors	All error codes explained
From Python	Python → Axon migration guide
From PyTorch	PyTorch → Axon migration guide

Tutorials

1. Hello, Tensor! — tensor basics
2. Linear Regression — autograd from scratch
3. MNIST Classifier — CNN training
4. Transformer — build a transformer encoder

---

Contributing

We welcome contributions! See the Contributing Guide for details on:

• Building from source
• Running and writing tests
• Code style guidelines
• Pull request process

# Quick start for contributors
git clone https://github.com/axon-lang/axon.git
cd axon
cargo build
cargo test

---

License

This project is licensed under the MIT License. See LICENSE for details.